Google is trying to make marketing measurement behave less like a collection of reports and more like an operating system for investment decisions. Its latest updates connect three pieces that marketers have traditionally handled separately: first-party data plumbing, marketing mix modeling and causal experimentation. AI is being added not merely to summarize the output, but to help teams diagnose data problems and build the models themselves.
In a September 10 announcement, Google said Meridian, its open-source Marketing Mix Model, is gaining agentic capabilities that can audit data quality, resolve errors and guide model construction in real time. Search Engine Land highlighted the broader rollout, which also brings Data Manager into Google Analytics and Display & Video 360, lets Meridian incorporate brand-demand signals such as Branded Google Query Volume, and makes Meridian GeoX generally available worldwide.
The common thread is an attempt to reduce the distance between collecting marketing data and deciding where the next dollar should go. Data Manager supplies and diagnoses first-party signals, Meridian estimates how channels contribute to business outcomes, and GeoX can inject evidence from geographic experiments into those models. The new agentic layer is intended to make that workflow less dependent on specialists manually debugging every dataset and modeling step.
Meridian’s AI is aimed at the difficult work before the model produces an answer
Marketing mix modeling has become attractive as user-level attribution becomes less complete, but MMM is not a push-button replacement for attribution. Model results depend on input quality, variable selection, assumptions, calibration and the ability to identify errors before they contaminate an investment recommendation. Google says Meridian’s new agentic capabilities will help audit that data, troubleshoot problems and provide guidance during model building, while backend changes are intended to make analysis faster and more efficient.
That is a more consequential use of AI than simply generating a narrative around a completed dashboard. If the tools work as intended, an agent can assist at the stage where analysts currently spend time checking whether a dataset is suitable, identifying configuration problems and navigating the technical decisions required to construct a model. It does not eliminate the need for measurement judgment: an AI assistant can help execute and diagnose a workflow, but the validity of the resulting business interpretation still depends on the model, assumptions and evidence available.
Google is simultaneously broadening what Meridian can represent by allowing relevant brand signals, including Branded Google Query Volume, to enter the model. The goal is to capture effects that occur between upper-funnel exposure and a final conversion. A television or video campaign may increase brand interest before sales materialize; branded search demand can provide an intermediate signal that helps a model represent that path rather than evaluating the campaign only through immediate conversions.
Branded search becomes a measurement signal, but not automatic proof of causality
The addition is notable for search marketers because branded queries have long sat ambiguously between outcome and attribution signal. A rise in searches for a company can indicate stronger demand, but the query volume alone does not establish what caused that demand. Seasonality, news coverage, offline distribution, competitor activity and other marketing can all move branded search at the same time. Meridian’s ability to incorporate the signal therefore makes it more useful for modeling brand effects, not inherently causal on its own.
Google’s architecture increasingly pairs that modeled evidence with experiments. Meridian GeoX, introduced in beta earlier this year, is now generally available globally. The open-source library is designed to run geographic incrementality tests across advertising platforms, comparing appropriately constructed test and control geographies to estimate whether an intervention actually caused additional outcomes. Those results can then be incorporated into Meridian to calibrate the marketing mix model against experimental evidence.
This distinction is central to Google’s measurement pitch. MMM can estimate relationships across a broad media mix, including channels where deterministic attribution is incomplete, while a well-designed incrementality experiment can provide stronger causal evidence for a specific intervention. Feeding experimental results back into the model is intended to constrain or calibrate estimates that otherwise rely more heavily on observational patterns and priors.
Data Manager is expanding from Ads into the wider measurement stack
The other half of the update concerns the data foundation feeding those systems. Google is directly integrating Data Manager into Google Analytics and Display & Video 360, expanding a workflow that previously centered more heavily on Google Ads. The Data Manager API is also becoming universal through the IAB Tech Lab’s Event and Conversions API standard, while built-in diagnostics are designed to identify and address data issues before they affect campaign performance.
For measurement teams, the significance is less about another connector and more about consistency. If first-party, offline and app signals can be managed through a common layer across Ads, Analytics and DV360, Google can reduce the configuration differences that make cross-platform measurement difficult. A new Data Strength Uplift Metric in Google Ads is intended to quantify additional conversions recovered through a company’s first-party data setup, giving advertisers another way to evaluate whether improving that foundation produces measurable value.
The direction has been visible throughout 2026. In May, Google outlined a measurement strategy built around stronger data connections, Meridian, GeoX and tools for scaling model creation. The September rollout turns several of those announcements into generally available or more deeply integrated capabilities and makes the relationship among them clearer: collect better signals, model the complete media mix, test causality where possible and use AI to reduce the technical work required to keep the system running.
None of this makes measurement automatic. Branded search is still a signal rather than causal proof, an MMM is still sensitive to its assumptions, and a geo-experiment is only as credible as its design. What Google is changing is the workflow around those limitations. Instead of asking marketers to choose between attribution, modeling and experimentation, it is building a stack in which first-party data, brand-demand indicators and causal tests can inform the same decision process.
That makes the “agentic” part of Meridian more than an AI feature announcement. Google is positioning agents as the connective tissue between messy marketing data and increasingly sophisticated measurement methods, while GeoX supplies a route to experimental calibration and branded query volume helps represent demand that occurs before conversion. The larger bet is that the next generation of marketing measurement will not be one perfect attribution model, but a system that continuously combines data quality, modeled relationships and causal evidence to decide what actually deserves more budget.